pytorch torchviz visualización de modelo de aprendizaje profundo

Descripción del problema

Al escribir un modelo de red neuronal, a menudo es imposible juzgar intuitivamente si el modelo que diseñamos es lo que imaginamos, por lo que es necesario visualizar el modelo.

solución:

1. Instale Torchviz:

pip install torchviz

2. Ejemplos de aplicación

Por ejemplo, hay una red neuronal de dos capas

class CNN2(nn.Module):  #3 layers
    def __init__(self,num_classes=10):
        super(CNN2, self).__init__()

        self.layer1 = nn.Sequential(
            nn.Conv2d(1, 16, kernel_size=3),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        self.layer2 = nn.Sequential(
            nn.Conv2d(16, 16, kernel_size=1),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        

        
        self.fc1 = nn.Linear(1*10816, 2048)
        self.fc2 = nn.Linear(2048, num_classes)

    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = x.view(x.size(0), -1)
        x = self.fc1(x)
        x = self.fc2(x)

        return F.log_softmax(x, dim=1)

Funciones personalizadas para imprimir parámetros del modelo y visualizar imágenes

def modeltorchviz(model,input,input2):
    # from torchviz import make_dot
    # params: model = MSDNet(args).cuda()
    # params: input = (3, 32, 32) 
    # params: input2 = torch.randn(1, 1, 28, 28).requires_grad_(True)  # 定义一个网络的输入值
    print(model)        
    summary(model, input)
    y = model(input2.cuda())    # 获取网络的预测值
    MyConvNetVis = make_dot(y, params=dict(list(model.named_parameters()) + [('x', input2)]))
    MyConvNetVis.format = "png"
    # 指定文件生成的文件夹
    MyConvNetVis.directory = "data"
    # 生成文件
    MyConvNetVis.view() 

definir la función principal

def main():
    model = CNN2(10).cuda()
    input = (1,28,28)
    input2 = torch.randn(1, 1, 28,28).requires_grad_(True)  # 定义一个网络的输入值
    modeltorchviz(model,input,input2)
    
  
if __name__ == '__main__':
    main()

El resultado es el que se muestra en la imagen



¡Éxito!
código completo

import torch
from torch import nn
import torch.nn.functional as F
from torchsummary import summary
import hiddenlayer as h
from torchviz import make_dot

class CNN2(nn.Module):  #3 layers
    def __init__(self,num_classes=10):
        super(CNN2, self).__init__()

        self.layer1 = nn.Sequential(
            nn.Conv2d(1, 16, kernel_size=3),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        self.layer2 = nn.Sequential(
            nn.Conv2d(16, 16, kernel_size=1),  # 16, 26 ,26
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True)
        )
        

        
        self.fc1 = nn.Linear(1*10816, 2048)
        self.fc2 = nn.Linear(2048, num_classes)

    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = x.view(x.size(0), -1)
        x = self.fc1(x)
        x = self.fc2(x)

        return F.log_softmax(x, dim=1)

def modeltorchviz(model,input,input2):
    # from torchviz import make_dot
    # params: model = MSDNet(args).cuda()
    # params: input = (3, 32, 32) 
    # params: input2 = torch.randn(1, 1, 28, 28).requires_grad_(True)  # 定义一个网络的输入值
    print(model)        
    summary(model, input)
    y = model(input2.cuda())    # 获取网络的预测值
    MyConvNetVis = make_dot(y, params=dict(list(model.named_parameters()) + [('x', input2)]))
    MyConvNetVis.format = "png"
    # 指定文件生成的文件夹
    MyConvNetVis.directory = "data"
    # 生成文件
    MyConvNetVis.view() 
    
def main():   
    model = CNN2(10).cuda()
    input = (1,28,28)
    input2 = torch.randn(1, 1, 28,28).requires_grad_(True)  # 定义一个网络的输入值
    modeltorchviz(model,input,input2)
    
  
if __name__ == '__main__':
    main()